Science1 publisher2 min readPublished
OpenAI's claimed proof of a famous math problem starts a fight over authorship and standards
OpenAI said on September 8 that its models had solved one of mathematics' most famous problems, setting off a dispute over authorship and proof standards. The fight now turns on whether proofs that mathematicians can barely read can be checked and credited at all.
The Scientist · Science desk
Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction
What happened
- Through the summer of 2026, AI models were putting out proofs of decades-old conjectures seemingly every week, according to Quanta Magazine.
- Scott Aaronson of the University of Texas, Austin wrote that human mathematicians are "forevermore dethroned" as the main theorem-proving entities on the planet.
- Two days after the announcement, Ken Ono, a mathematician who left academia for start-up Axiom Math, spoke to some 150 students, postdocs and professors at UC Berkeley.
- Quanta's writer says current AI proofs elide salient details, spend pages on irrelevant concepts and fail to show how they connect to existing results.
Compiled by The ScientistSomething wrong?How this is made
Why it matters
- cost Referees and journal editors pay for this: every machine proof put forward for acceptance still needs human hours to follow, and a proof that is hard to read needs more of them.
- decision Journals, departments and prize committees will have to state whether a model can be credited with a proof, and who answers for an error in one.
- exposure The students and postdocs Ono addressed are choosing their training now for a job whose content is in question, so they carry most of the career risk.
Quanta's account does not name the problem OpenAI says its models solved [4]. With no independent check of the work reported, and no department or journal rule yet on crediting machine-written proofs, the dispute that followed, over authorship and standards [5], is so far an argument about credit and process.
The most concrete complaint in the piece is about reading. Quanta's writer says AI proofs are, at the moment, nearly impossible to read [1]. I think that is where the standards question will be decided. Checking a proof by hand means following it step by step, and these proofs are hard to follow. Each proof in the weekly tally of proved conjectures [14] still takes human time to confirm.
The two academics quoted on what is being lost disagree about what it is. Scott Aaronson wrote on his blog: "In whatever years I have left, I don't expect that I'll ever again prove a theorem because I'm actually needed to prove it." [6] Marcel Goh, a doctoral student at McGill University, located the value somewhere else. "It was never about the theorems," Goh said. "It's to carry on a tradition that has benefited society despite not having any market value." [12]
Quanta's writer describes the machines plainly: "AI models skip to the end." [3]
At Berkeley the argument got personal. "You might be graduating into a profession that might not even exist, or that will be very different than what you expected," Ken Ono told the audience. "You need to brace." [9] The talk had a 50-minute slot and, with questions, ran more than two hours [10], more than 2.4 times its allotted length [13]. When Ono said mathematicians would do their "very best" to avoid the future the students feared, one of them replied: "What are you doing? What is your 'very best'?" [11]
Quanta's writer argues that mathematicians now have to settle for themselves what mathematics is and why they do it, explain that to the rest of the world, and do it quickly. In the writer's view, the answer will determine what happens next [15].
What to watch
- Whether OpenAI names the problem and releases the proof in a form outside specialists can check, and whether one of them confirms it.
- Whether any journal or mathematics department publishes a rule on listing AI systems as authors or on refereeing machine-generated proofs.
- Whether newer AI proofs start showing how they connect to the existing literature, the readability problem Quanta describes in current models.